- Python 3.5
- Tensorflow 1.3.0
- NumPy 1.13.3
- DLIB 19.7.0
- matplotlib 2.1.3
Run python3 train.py [data_directory]
to generate cropped data and train a model.
The data will be saved to data.npz
. Next time, the training will start immediately. Delete data.npz
to generate it again.
data_directory
must contain subdirectories (300W
and Menpo
) with train
and test
folders in them.
Run python3 test.py
to launch a demonstration of the algorithm.